Assemblyai Speech

A speech interface needs to know when a speaker has finished, not merely when text has been prettified. AssemblyAI's turn-detection documentation distinguishes end_of_turn from turn_is_formatted.

For the streaming speech entry, retain turn identifiers and handle those events independently. Test a mid-sentence pause followed by more speech. The UI should update the same turn when formatting arrives, without starting a duplicate response or treating every pause as a completed request.